.. | ||
tasks-solutions | ||
README.md |
Threads and Tasks
A satellite is a complex system which usually has a lot of tasks which need to be done simulatenously by a dedicated On-Board Computer (OBC). This can include for example:
- TMTC handling. This includes Telecommand (TC) reception and execution, and the (autonomous) generation of Telemetry (TM)
- Control Operations, for example execution of the Attitue Control System (ACS) loop
- Handling of connected physical devices like sensors or payloads
Usually, these systems oftentimes have soft and even hard real-time requirements where longer delays are not allowed and the system has an upper bound for response times.
This basically means that any software which does multiple non-trivial tasks needs a (real-time) operating system to perform multiple tasks consecutively, with deterministisc guarantees that these tasks are performed within a certain temporal bound.
Some common operating system in the Space domain able to do this:
- FreeRTOS for smaller MCUs (e.g. SOURCE CubeSat project)
- Embedded Linux (EIVE CubeSat project)
- RTEMS (FLP satellite project)
All these operating system use threads or tasks as the basic worker unit which is executing code. This chapter first introduces threads as they are exposed by the C++ standard library. After that, the code is transitioned to use the abstraction provided by the framework.
1. Scheduling a basic task using the C++ std::thread
API
The goal of this task is to set up a basic thread which prints the following string every second: "Hello World".
2. Changing to the concept of executable objects
The goal of this task is to convert the code from task 1 so the std::thread
API takes an
executable object to move to a more object oriented task approach. The printout of the thread
should remain the same. The executable objects should be named MyExecutableObject
. It contains
one function called periodicOperation
which performs the printout, and a static function which
takes the MyExecutableObject
itself by reference and executes it in a permanent loop.
The executable object should be passed into the std::thread
directly.
Hints
- std::reference_wrapper to pass references to the [std::thread] API
- std::chrono::milliseconds has a constructor
where an
uint32_t
can be used to create the duration from a custon number.
Subtasks
- Create a class called
MyExecutableObject
with apublic
block. - Add a static function called
executeTask
which expects itself (MyExecutableObject& self
) as a parameter with an empty implementation - Add a regular method called
performOperation
which performs the printout - Implement
executeTask
. This function uses the passed object and performs the scheduling specific part by callingself.performOperation
in a permanent loop with a delay between calls. You can hardcode the delay to 1000ms for the first implementation. - Add a constructor to
MyExecutableObject
which expects a millisecond delay as anuint32_t
and cache it as a member variable. Then use this member variable in theexecuteTask
implementation to make the task frequency configurable via the constructor (ctor) parameter.
With the conversion to executable object, we have reached a useful goal in object-oriented
programming (OOP) in general: The application logic inside performOperation
is now decoupled
from the scheduling logic inside executeTask
. This is also called seperation of concerns.
3. Making the executable objects generic
Threads generally expect a function which is then directly executed. Sometimes, the execution of threads needs to be deferred. For example, this can be useful if the execution of tasks should only start after a certain condition.
Also, it might become useful to model any task in form of a class. An instantiation of that class would then be an executable object. Another point is that even though we have seperated the scheduling specific part from the application logic, they are still part of the same class. It would be nice to have two separate classes for this.
C++ as an OOP language provides abstraction in form of interfaces, which can be used to have different types of generic executable objects. Interfaces usually do not have a lot of source code on their own. They describe a design contract a class should have which implements the interface. In general, the FSFW relies heavily on subclassing and inheritance to provide adaptions point to users.
We are going to refactor our MyExecutableObject
by introducing an interface for any executable
object. We are then going to add a generic class which expects an object fulfilling this design
contract and then executes that object.
Interfaces in C++ are implemented using abstract classes which only contains pure virtual functions.
Subtasks
-
Create an interface called
MyExecutableObjectIF
. You can create this like a regular class. As opposed to Java the differences between interfaces and classes are only by convention. -
In general, it is recommended to add a virtual destructor to an interface. It looks like this:
virtual ~<Class>() = default;
-
Add a abstract virtual function
performOperation
. Abstract virtual functions look like this in generalvirtual <functionName>(...) = 0;
-
Implement you custom interface for
MyExecutableObject
by re-using the exsitingperformOperation
function. In general, when implementing an interface or overriding a virtual function, it is recommended to add theoverride
keyword to the function delaration. We do not have seperation between source and header files for our class yet, so you can add theoverride
keyword after the function arguments and before the implementation block. The compiler will throw a compile error if a function is declared override but no base object function was actually overriden. This can prevent subtle bugs. Please note thatMyExecutableObject
is actually now forced to implement theperformOperation
function because that function is pure. The compiler makes sure we fulfill the design contract specified by the interface -
Add a new class called
MyPeriodicTask
. Our executed object and the task abstraction are now explicitely decoupled by using composition. Composition means that we have a "has-a" relationship instead of a "is-a" relationship. In general, composition is preferable to inheritance for flexible software designs. The newMyPeriodicTask
class should have a ctor which expects aMyExecutableObjectIF
by reference and the task frequency in milliseconds as anuint32_t
. It caches both the executbale object and the task frequency as private member variables. -
Add a public
start
function which returns astd::thread
and leave it empty for now. -
Add a private static
executeTask
method which expectsMyPeriodicTask
by reference. Its implementation is similar to theexecuteTask
method ofMyExecutableObject
. Remove theexecuteTask
implementation fromMyExecutableObject
. -
In the start method, use
std::thread
API withMyPeriodicTask::executeTask
as the executed function. Pass the task itself by reference similarly to how it was done in task 2. Return the created thread directly, so callers can use thejoin
method to block on thread completion.
We now have two separate classes where one class only contains application logic and the other
one only contains scheduling logic. The MyPeriodicTask
is also able to schedule arbitrary
types which implement MyExecutableObjectIF
. Finish this task
by crating 3 different executable objects where each object does or prints our something
different. Then pass all of those three different objects to a MyPeriodicTask
and start
all three periodic tasks which three different frequencies.
You now should have code which looks something like this:
int main() {
MyExecutableObject0 myExecutableObject0;
MyExecutableObject1 myExecutableObject1;
MyExecutableObject2 myExecutableObject2;
MyPeriodicTask task0(myExecutableObject0, 1000);
MyPeriodicTask task1(myExecutableObject1, 2000);
MyPeriodicTask task2(myExecutableObject2, 5000);
auto thread0 = task0.start();
auto thread1 = task1.start();
auto thread2 = task2.start();
thread0.join();
thread1.join();
thread2.join();
return 0;
}
Where the three tasks do their tasks with different frequencies.
4. Using the framework abstractions
We now use framework components to perform the tasks shown above. The framework
exposes an abstractions for executable tasks called ExecutableObjectIF
.
It also offers a unform API to execute periodic tasks in form of the
PeriodicTaskIF
.
These tasks can then be created using the
TaskFactory
singleton.
An arbitrary number of executable objects can then be passed to a periodic task. These objects are then executed sequentially. This allows a granular design of executable tasks. For example, important tasks get an own dedicated thread while other low priority objects are scheduled consecutively in another thread.
In summary, task abstractions have the following advantages:
- Task execution can be deferred until an explicit
start
method is called - Same uniform API across multiple operating systems
Subtasks
- Load the required interfaces:
#include "fsfw/tasks/ExecutableObjectIF.h"
#include "fsfw/tasks/PeriodicTaskIF.h
#include "fsfw/tasks/TaskFactory.h
- For your three custom objects, implement the executable object IF provided by the framework instead of your custom interface.
- In your main function, create an instance of the
TaskFactory
.TaskFactory
is implemented as a singleton: The object will create itself when usingTaskFactory::instance
. All subsequent calls will return the same intance. There are other similar singletons to create other objects like mutexes or message queues. - Create two periodic tasks using the
TaskFactory::createPeriodicTask
function. Some notes on the expected arguments:- Each task has a name. This is useful for debugging, especially because the framework abstractons can detect missed deadlines.
- The task priority parameter is OS dependent. This parameter is currently ignored for the
Linux OSAL so you can pass 0 here.
On Windows, you can retrieve the priority by using
tasks::makeWinPriority
which can be loaded by including#include "fsfw/osal/windows/winTaskHelpers.h"
- The stack space parameter is generally ignored or unimportant for host systems.
You can simply pass
PeriodicTaskIF::MINIMUM_STACK_SIZE
here. This parameter becomes important on resource constrained OSes and systems, for example FreeRTOS. - The frequency is expected as floating point seconds
- You can pass a function which will be called when a deadline is missed. This is the case
when a tasks took longer than its designated slot frequency. This is useful to detect
bugs in the software or generally detect when tasks require a large amount of time.
You can also pass
nullptr
here if you do not want any function to be called.
- Add the first two of your custom exec objects to the first periodic task. Those tasks
will be executed in the same thread consecutively. You can use the
PeriodicTaskIF::addComponent
method to do this. - Add the third custom exec object to the second periodic task. The third object gets an own thread
- Start both periodic tasks and add a permanent loop at the end of the main method which puts the main thread into sleep.
You successfully scheduled some objects using the framework!
The general concept of executble objects is used heavily throughout the framework.
For example, each device handler or controller is an executable object, as the bases
classes exposed by the framework implement ExecutableObjectIF
.
There is also another type of periodic task handler called FixedTimeslotTask
. Here,
you can explicitely specify (multiple) execution slots with a specified relative time within
the execution slot. This is useful for objects where there are multiple processing steps
but the steps take different amount of times.
In examples or other OBSW implementations using the framework, you will often
see the distinction between an ObjectFactory.cpp
and an InitMission.cpp
.
In the first file, all global (executable) objects will be created. In the second file,
all of these objects will be scheduled. Another chapter will introduce the Object Manager
to show what exactly is happening here.